{"id":"84d78bf2-7f74-48e1-bc6c-77c43ca4828a","arxiv_id":"2611.17069","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":1,"one_line_summary":"The median angular error of an incoherent pulsar timing array scales as sqrt(ln 2/3) / (SNR sqrt(ln SNR)), about 3.3 degrees at SNR 6, which is 1.4 times better than a Gaussian beam.","lead":"This paper derives a scaling relation for how precisely a pulsar timing array can locate a gravitational wave source on the sky when pulsar distances are unknown. It shows that the angular error shrinks faster with signal strength than previously expected, and quantifies the actual performance of the International Pulsar Timing Array.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"The Fisher-information cutoff at epsilon ~ gamma is a heuristic that may not capture the true small-angle behavior; the scaling pre-factor could shift if the cutoff is resolved differently.","rationale":"The reader's weakest_assumption correctly identifies the cutoff at epsilon ~ gamma as a heuristic approximation. However, the reader does not flag this as a load-bearing concern for the central claim; they note it as an assumption but still accept the paper. I agree that the paper is well-supported overall, but I think the cutoff is the single most load-bearing point because the quantitative prefactor (and thus the claimed 1.4x improvement) depends on it. The simulation shows consistency, but the simulation itself uses the same heuristic implicitly? No, the simulation is a direct Monte Carlo, so it does not use the cutoff; it is an independent check. The agreement between simulation and Eq. 12 suggests the heuristic is reasonable. However, the simulation is only for one array configuration and one SNR range. A self-consistent derivation would strengthen the claim. I recommend UNCHANGED because the paper already provides simulation support, and the concern is about rigor of the analytic derivation, not about correctness of the result. The concrete test would settle whether the heuristic cutoff is accurate, but the paper's central claim is already backed by simulation, so the verdict should not change. Agreement with reader is partial: the reader identified the same assumption but did not consider it load-bearing; I consider it the most load-bearing but still not sufficient to change the verdict.","tokens_in":13184,"tokens_out":748,"duration_ms":9168,"concrete_test":"Re-derive the Fisher information integral (Eq. 11) without the heuristic cutoff by computing the exact response derivatives for a source at the pole and integrating over the full sphere, then solving the self-consistent equation gamma_med = sqrt(ln 2 / F(gamma_med)). Compare the resulting gamma_med to Eq. 12 at SNR = 6 and 10. If the self-consistent solution differs by more than 10%, the heuristic cutoff is the source of the discrepancy.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The derivation of the scaling relation (Eq. 12) relies on a logarithmic integral that is cut off at epsilon ~ gamma, where gamma is the localization error itself. This is a self-consistency condition rather than a first-principles cutoff: pulsars closer than the localization error are said to be 'fully de-phased,' but the Fisher information integral (Eq. 11) is evaluated assuming the response derivatives are still given by the small-angle expansion. If the true response saturates or behaves differently at angular separations comparable to or smaller than gamma, the prefactor and possibly the logarithmic scaling could change. The authors acknowledge this is a heuristic (sentence b78: 'The integral diverges logarithmically at small theta and is cut off at epsilon ~ gamma, since pulsars closer to the source than the localization error are fully de-phased.'). The simulation in Fig. 1 shows agreement with Eq. 12 at high SNR, but the test is performed for a single array configuration (4800 uniform pulsars) and a single source placement (equator, azimuth randomized). The cutoff prescription is not independently verified; the simulation could be consistent with a different effective cutoff that yields the same functional form but a different prefactor. Because the central claim is the specific scaling and prefactor (the 1.4x improvement over Gaussian), the uncertainty in the cutoff directly affects the quantitative prediction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper derives and verifies a scaling relation for the astrometric precision of an incoherent pulsar timing array (PTA), showing that the median angular error scales as sqrt(ln 2/3) / (SNR sqrt(ln SNR)) at high SNR, which is sharper than the 1/SNR scaling of a Gaussian beam. The authors use a Fisher information calculation with a heuristic cutoff at epsilon ~ gamma, and confirm the relation with simulations of a uniform array of 4800 pulsars. They then compare the intrinsic precision (3.3 deg at SNR=6) with the precision achievable with the IPTA DR2 pulsar distribution (median 8.9 deg, but 14% of sky better than 3.3 deg).","tokens_in":13600,"tokens_out":1998,"duration_ms":149554,"significance":"If correct, the result is significant because it identifies a fundamental scaling that improves localization precision beyond the naive Gaussian-beam expectation, which is relevant for planning future PTA observations and for interpreting the localization capabilities of current arrays. The paper provides a clear derivation and simulation verification, and it quantifies the performance gap between an ideal uniform array and the actual IPTA configuration.","major_comments":[{"comment":"The Fisher information integral is cut off at epsilon ~ gamma, where gamma is the localization error itself. This is a self-consistency condition rather than a first-principles cutoff. The authors acknowledge this is heuristic (sentence b78). However, the central quantitative prediction—the prefactor in Eq. (12) and the claimed 1.4x improvement over Gaussian—depends on this cutoff prescription. The simulation in Fig. 1 tests only one array configuration (4800 uniform pulsars) and one source placement (equator, randomized azimuth). The agreement with Eq. (12) at high SNR is encouraging, but it does not independently verify the cutoff; a different effective cutoff could yield the same functional form with a different prefactor. The authors should either (a) provide a more rigorous justification for the cutoff, (b) test the sensitivity of the prefactor to the cutoff choice (e.g., by varying the cutoff prescription in the Fisher integral and comparing to simulations), or (c) explicitly state the uncertainty in the prefactor due to this heuristic. Without this, the quantitative claim of 1.4x improvement is not fully supported.","section":"§III A, Eq. (11) and surrounding text","authors_acknowledge":"(11) The integral diverges logarithmically at smallθand is cut off atϵ∼γ, since pulsars closer to the source than the localization error are fully de-phased.","authors_acknowledge_p":0.64},{"comment":"The simulation uses 500 noise realizations at each SNR, but the same 500 realizations are reused across SNR values by rescaling the signal amplitude. This is acceptable for studying the scaling, but the error bars on the median angular error are not shown. The paper should include error bars (e.g., bootstrap confidence intervals) on the simulation points in Fig. 1 to assess the statistical uncertainty and confirm that the agreement with Eq. (12) is significant. Without error bars, the reader cannot judge whether the small deviations at high SNR are consistent with the model.","section":"§III B, Fig. 1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The IPTA DR2 simulation uses N_real = 200 noise realizations per source position. The median angular error map (Fig. 2) and the cumulative distribution (Fig. 3) are based on these 200 realizations. The authors should report the uncertainty on the median precision (e.g., via bootstrap) and on the 14% sky fraction. Given the small number of realizations per pixel, the median may be noisy, especially in regions with low pulsar density. The paper should discuss the robustness of the 14% figure.","section":"§IV, Fig. 2 and Fig. 3","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The Fisher information expression includes a factor of 3/8 SNR^2. The derivation of this prefactor should be clarified: the text says 'normalizing with the SNR definition of Eq. (6), for which sum_i |zeta_i|^2 = 4N/3'. It would be helpful to show the intermediate steps or provide a reference to a more detailed derivation.","section":"§III A, Eq. (11)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The final expression gives gamma_med in radians, then converts to degrees: 27.5 deg / (SNR sqrt(ln SNR)). The constant 27.5 deg comes from sqrt(ln 2/3) * (180/pi). This is correct, but the paper should double-check the arithmetic: sqrt(ln 2/3) ≈ sqrt(0.231) ≈ 0.481, times 180/pi ≈ 57.3 gives 27.6, consistent. However, the text says 'rad≈ 27.5°' which is fine.","section":"§III A, Eq. (12)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The figure caption says 'Lines show the Gaussian-PSF scaling γ_{1/2}/SNR with γ_{1/2}=28.9° [7], and the derived relation ... of Eq. (12).' It would be helpful to also plot the asymptotic scaling of Eq. (12) without the prefactor (i.e., just 1/(SNR sqrt(ln SNR))) to show that the functional form is verified, not just the prefactor.","section":"§III B, Fig. 1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The color scale for the median angular error map is not specified in the caption. The figure should include a color bar with units (degrees). Also, the discontinuities mentioned in the text (near the ring of pulsars) are not clearly visible in the grayscale version; consider using a perceptually uniform colormap.","section":"§IV, Fig. 2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper states that 'the configuration-dependent gains outlined in Ref. [9] could become possible to leverage once a candidate source has been identified.' This is an interesting point, but it would be strengthened by a brief quantitative estimate of how much improvement could be expected for a known source direction, given the IPTA configuration.","section":"§V, Discussion","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper uses the term 'astrometric precision' to mean the median angular error of the maximum-pixel estimator. This is fine, but it should be explicitly distinguished from other measures like the Cramer-Rao bound or the RMS error, especially since the distribution of angular errors has a long tail at low SNR. The authors already note that they use the median because it is more robust, but they should also mention that the RMS error would be larger.","section":"General","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The dirty map is defined as L = (A^L)^† Σ^{-1} d. For complex data, the Hermitian conjugate should be used. The paper states that the noise is complex Gaussian, so the map is complex. This is fine, but the notation should be consistent: (A^L)^† is the conjugate transpose.","section":"§II, Eq. (3)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The heuristic argument dividing the response map into SNR^2 equal-area rings is clever, but the derivation of Eq. (9) uses the approximation that the ring SNR is constant (SNR_k = 1). The authors note that this approximation only changes the constant, not the scaling. However, the constant does affect the prefactor in Eq. (12). It would be helpful to show that the Fisher information derivation (Eq. 11) gives the same prefactor as the heuristic argument, or to note that the heuristic argument is only used to motivate the scaling, while the prefactor comes from the Fisher calculation.","section":"§III A, heuristic argument","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The cumulative distribution function in Fig. 3 shows the fraction of sky with median angular error better than a given value. The vertical dashed lines mark the IPTA sky median (8.9°) and the intrinsic precision (3.3°). The horizontal line marks 14%. The figure would be clearer if the axes were labeled with units and if the curves were distinguished by line style or color.","section":"§IV, Fig. 3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The conclusion restates the main results. It might be useful to add a sentence about the implications for future PTA design: e.g., that increasing the number of pulsars and their sky coverage is important for improving localization, and that the scaling relation provides a target for the best possible precision.","section":"§VI, Conclusion","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Colleague,\n\nThis paper derives the scaling of astrometric precision for an incoherent PTA: median angular error ~ 1/(SNR sqrt(ln SNR)), with a prefactor. That is genuinely new and useful. The derivation is clear, the simulation (4800 uniform pulsars, 500 noise realizations) matches the formula at high SNR, and the comparison with IPTA DR2 gives a concrete sense of where current arrays stand. The paper does what it sets out to do.\n\nThe main soft spot is the cutoff at epsilon ~ gamma in the Fisher integral. The authors call it a heuristic, and it is. The integral diverges at small theta, and they cut it off at the localization error itself, which is a self-consistency condition rather than a first-principles resolution of the singularity. The stress-test note worries that a different cutoff could shift the prefactor. That is a legitimate concern in principle, but in practice the simulation agrees with the derived prefactor at high SNR for the one configuration tested. The prefactor could shift by an O(1) factor if the cutoff were resolved differently, but the functional form (the 1/(SNR sqrt(ln SNR)) scaling) is robust. The 1.4x improvement over Gaussian is a modest quantitative claim, not a dramatic one, so even a 20% shift in the prefactor would not change the paper's message.\n\nA minor point: the simulation uses only one array configuration (4800 uniform pulsars) and one source placement (equator, randomized azimuth). The scaling is verified only for that case. The authors acknowledge this implicitly by calling it the \"intrinsic\" precision, but a reader might want to see that the scaling holds for other dense configurations or for sources at different latitudes. Not a flaw, just a limitation of the current test.\n\nThe paper is written for PTA practitioners who want to know how well they can localize a single source. It is a solid, focused contribution. I would send it to a serious referee.\n\nBest,\n[Your name]","headline":"A clean derivation of PTA astrometric precision scaling, with a heuristic cutoff that probably doesn't change the answer much.","tokens_in":14075,"tokens_out":508,"duration_ms":16452},"pith_extraction":null,"created_at":"2026-11-13T10:26:13.980063+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":null,"supporting_citations":[],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}